Reduce rendering cost in long chats (#7052)

* perf(app): stop re-walking the timeline on every streamed delta

Every coalesced delta rebuilds the stream render model with the same
tail. The model re-sliced the tail at the history-window start on every
call, so the rendered tail was a fresh array per delta and every WeakMap
keyed on it (ordering, history split, turn timing, the layout history
cache, row revision) missed each time. Cache the slice by tail identity
and window start.

Turn timing then still walked the whole rendered tail per delta and
allocated a new assistant-timing Map, which the layout memo depends on.
Split the walk: the tail is walked once per tail identity and the head is
walked from the carried open turn per delta. The Map keeps its identity
while the head adds no completed turn.

Adds identity-contract tests and a vitest bench for the per-delta
pipeline. 200 deltas over a 2,000-item timeline, mean per run:
default mounted window 5.68ms -> 2.82ms, history fully revealed
16.95ms -> 1.81ms.

Fixes #4816

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* perf(app): keep the stream view's per-delta work bounded by the head

The chat outline concatenated tail and head, reduced over the result in
the render body, and rebuilt jumpToPrompt on every delta even when the
outline was disabled, which is every native viewport. Memoize the tail's
contribution, scan only the head per delta, do nothing while disabled,
make jumpToPrompt a stable event so the rail's memo holds, and republish
the active prompt only when the index or reading position changes.

The view built a Set of every mounted id plus the live head per delta for
the outline's pending-jump check; key it on mounted history only and
treat live head rows as always mounted. Split plugin projection so a head
change does not re-project the tail, memoize the history-window boundary
lookup, and read the turn presentation's submission flag with some()
instead of allocating a filtered array.

Refs #4816

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

* test(app): confirm full history before measuring streaming

---------

Co-authored-by: Ivan Malison <IvanMalison@gmail.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: paseo-bot[bot] <266920839+paseo-bot[bot]@users.noreply.github.com>

docs/agent-stream-performance.md

@@ -58,12 +58,17 @@ So arrival sets a _target_ and the reveal rate is derived from the backlog inste
   through that hook; the web viewport once skipped it and history hosts of a live tool group went
   stale. A new field on `StreamLayoutItem` must be added to `areLayoutItemsEquivalent`, or sharing
   silently stops.
+- **Streaming cost must not grow with revealed history.** Reading older messages must not
+  make the next reply progressively more expensive to display. Reuse unchanged history work
+  while text arrives, and release obsolete window caches as the reader reveals more history;
+  window caches must stay bounded while their source timeline remains loaded.
 
 ## Measuring
 
 - **Smoothness (user-perceived):** `packages/app/e2e/browser/agent-stream-smoothness.spec.ts`, gated behind `PASEO_AGENT_STREAM_PERF_E2E=1`. Drives the mock provider's `bursty-stream` model and reports coefficient of variation of characters painted per frame (smoothness) plus p95 gap between visible updates (stalls). Both numbers are needed: a stalled stream is perfectly smooth.
 - **Reproducing bursty arrival:** the `bursty-stream` model in `mock-load-test-agent.ts` emits uneven runs of tokens separated by idle gaps. Burst sizes come from a seeded generator, so a run repeats exactly.
 - **Rate policy in isolation:** `computeRevealStep` in `packages/app/src/agent-stream/text-reveal.ts` is pure; `text-reveal.test.ts` covers convergence and burst flattening without a renderer.
+- **Per-delta cost on a long timeline:** `npx vitest bench --project unit src/agent-stream/stream-tick.bench.ts` from `packages/app` replays 200 deltas over a 2,000-item timeline through the pure model and layout pipeline, with the default mounted window and with history fully revealed. It measures the JS work per delta, not paint.
 
 Healthy numbers (2026-08, Expo web against a local dev daemon, real Claude Haiku agent, ~8.5s samples during active streaming). Setting `TEXT_REVEAL_HORIZON_MS` to 0 makes the reveal paint on arrival, which is how the baseline column was taken: